Ship the Tool-Using Agent

assembling and shipping the agent

Part of: Tool-Using Agent

Every piece is built: schemas, reading a request, execution, routing across tools, the loop, validation, and a hard cap with error recovery. This lesson wires all of it into one script you can run, and finishing it lands the build in your Portfolio. The finished shape That's the whole product. Every function it calls is a lesson you already built: safe execute, the three run tool implementations, and the three schema dicts. What "shipped" means for an agent Three checks, from the playbook: 1. It runs from a clean start with one command and a real question. 2. Weird input doesn't crash it. An unknown city, a malformed expression, a tool that errors: each comes back as a graceful message, not a stack trace. 3. Someone else could point it at their own API key and use it right away. Watching cost on a tool-using agent Every extra tool call is another full round trip through the API, system prompt and growing history included. MAX ITERATIONS is not only a guard against infinite loops. It is a cost ceiling per question. If real usage keeps hitting the cap, that is a sign to raise it deliberately or tighten your tool descriptions so the model needs fewer calls. Into your Portfolio Finishi

Challenge: Agent Run Report